obsidian-vault-mcp
An MCP server for Obsidian vaults that handles iCloud eviction gracefully, allowing reading, searching, creating, and updating notes without hanging.
README
<p align="center"> <img src="./assets/header.svg" alt="obsidian-icloud-mcp — an Obsidian MCP that survives iCloud eviction" width="100%"> </p>
<p align="center"> <a href="https://github.com/QEbellavita/obsidian-icloud-mcp/actions/workflows/test.yml"><img src="https://github.com/QEbellavita/obsidian-icloud-mcp/actions/workflows/test.yml/badge.svg" alt="tests"></a> <a href="https://doi.org/10.5281/zenodo.21485168"><img src="https://zenodo.org/badge/DOI/10.5281/zenodo.21485168.svg" alt="DOI"></a> </p>
An MCP server for Obsidian vaults that doesn't hang on iCloud.
Read, search, create and update notes from your assistant — and see exactly how much of the vault iCloud has actually left on disk. Seven tools, no daemon, no index to rebuild.
Why another one
Because the obvious implementation breaks on iCloud Drive, and breaks in the worst possible way: it hangs forever.
If your vault lives in iCloud with Optimize Mac Storage on, macOS evicts note bodies
and leaves dataless placeholders behind. They look like ordinary files — stat reports
a real size — but the first read() blocks while macOS fetches the content over the
network. On an evicted vault, a naive search hits hundreds of those in a row and never
returns. No error, no timeout, just a search that never finishes.
A deadline alone doesn't fix it: a single blocking read never yields control back, so the
between-files clock check is never reached. The only reliable signal is stat's block
count — a dataless placeholder reports blocks == 0 despite a non-zero size.
This server checks that before every read, and bounds total search time on top. Evicted notes are skipped and reported, not waited on:
{
"matches": [ ... ],
"scanned": 112,
"total": 486,
"unavailable": 43,
"incomplete": true,
"note": "Searched 112/486 notes; 43 are not downloaded from iCloud yet and were skipped. Results may be incomplete."
}
incomplete: true means "there may be more" — rather than silently returning partial
results as if they were the whole answer. The counts distinguish the two reasons: notes
that weren't materialised (unavailable) versus hitting the time budget.
Reading one specific evicted note fails honestly with "not downloaded from iCloud" instead of blocking.
<p align="center"><img src="./assets/demo.gif" alt="Animated demo — a search reports 5 of 8 notes scanned with 3 iCloud-evicted notes skipped and flagged, then vault_health shows materialized vs evicted counts and points at obsidian_warm_notes" width="90%"></p>
Install
Point your MCP client at it:
{
"mcpServers": {
"obsidian": {
"command": "npx",
"args": ["-y", "obsidian-icloud-mcp"],
"env": {
"OBSIDIAN_VAULTS": "{\"Personal\":\"/Users/you/Documents/Notes\"}"
}
}
}
}
Or from source: git clone https://github.com/QEbellavita/obsidian-icloud-mcp and point
command: node, args: [/absolute/path/to/obsidian-icloud-mcp/server.js] at the checkout.
Formerly named obsidian-vault-mcp; renamed because that npm name belongs to an unrelated
project by another author — npx obsidian-vault-mcp is not this server. The old GitHub URL
redirects here.
OBSIDIAN_VAULTS is a JSON object of name → absolute path. Multiple vaults are fine:
"OBSIDIAN_VAULTS": "{\"Personal\":\"/Users/you/Notes\",\"Work\":\"/Users/you/Work/vault\"}"
With no vaults configured the server starts and advertises zero tools, rather than crashing in a way your host reports as a broken server.
Tools
| Tool | |
|---|---|
obsidian_list_notes |
List Markdown notes, optionally under a subdirectory |
obsidian_search_notes |
Full-text search with path, line number and excerpt |
obsidian_read_note |
Read one note |
obsidian_create_note |
Create a note; refuses to overwrite |
obsidian_update_note |
Atomic overwrite of an existing note |
obsidian_vault_health |
Materialized vs evicted counts, bytes pending, largest evicted notes — stat-only, never blocks |
obsidian_warm_notes |
Opt-in background downloads for evicted notes, bounded per call |
Safety
The vault boundary is enforced, not assumed:
- Paths must be relative and stay inside the configured vault —
../is rejected - Symlinks that resolve outside the vault are rejected, not followed
- Only
.mdfiles are read or written createrefuses to clobber;updaterefuses to create- Writes are atomic (temp file + rename), so an interrupted write can't truncate a note
Directory walking skips anything that can't hold notes — .git, node_modules,
.obsidian — which also stops a vault that happens to be a source checkout from taking
minutes to list. And it never descends into a stalling bundle directory, where even
readdir can block indefinitely on iCloud.
Tests
npm test
23 tests. Nineteen cover the service directly — including every iCloud failure mode above,
driven through an injected fs so eviction is simulated deterministically rather than
depending on your actual sync state. Four drive the real server over a real stdio
transport.
Citing
If this server is useful in your workflow, cite it via the repo's CITATION.cff (GitHub's "Cite this repository" button), or the Zenodo DOI, 10.5281/zenodo.21485168 — a concept DOI that always resolves to the latest archived version.
Licence
MIT — see LICENSE.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.